first post on here, so here's a proper intro.
i'm frey. for the last few years i've been an early engineer at ai startups, building them from the ground up, first in san francisco and now in london.
my journey started with not understanding things and being too curious to leave them alone. that led me to parallel computing at @dask_dev, quantum computing at @GoogleQuantumAI, compilers at @LLVMFoundation, ci tooling at @jenkinsci, cloud security at @Oracle, ai bias research at @AvidMldb, ai security at @RealmLabsAI, and now ai agents at @OvermindLab.
outside of work, i'm usually travelling, and the first phrase i learn in every new country is how to ask if i can pet someone's dogπ
@nvidia everybody becoming an ai company means everybody running agents. we built overmind so you can watch, fix and fine-tune every one of them, and we just open-sourced the whole thing so anyone can π
https://t.co/kCIXiRNiy9
watching jensen on stage in berlin today was surreal. "everybody will become an ai company" sounds like a big claim, until you spend a week here and see it already happening, from two-person startups to some of the biggest names in industry. congrats @nvidia, this felt like europe's moment!!
what stood out to me most is how open it all is. open models, open tooling, and a real push for companies to own their ai instead of renting it. that is the future i want to be building in. also, the tiny @pollenrobotics microducks waddling around jensen's feet stole the whole show. i need one immediatelyyy
the exciting bit about open models isn't just that they're open. it's that they let us turn proprietary data into sovereign intelligence, tailored to you and running on infrastructure you control.
think linux, but for intelligence. small, specialised models powering the core services, working together as a system anyone can build on.
@NVIDIAAI is pushing this future forward. at @OvermindLab, we're working to bring it to the masses ππ
Weβre releasing more open models to help teams across Europe build AI with their own data, on infrastructure they control.
Announced at GTC Berlin, the models help developers, researchers and enterprises build specialized AI agents, train robots, design proteins and develop quantum computers.
See whatβs new: https://t.co/gXbVJfRn1b
in berlin this week for #nvidiagtc, representing @OvermindLab. the last few years were about building the biggest engine, and the next few will be won by the pit crew. come say hi if you are building agents π
Coinbase built this for their employees.
Overmind ships it for everyone.
Run /overmind finetune in:
β’ Claude Code
β’ Cursor
β’ Codex
Open source, and the model is yours.
bridging the software (observability, evals, code optimizer, data ingestion) Γ hardware (training, inference, hosting) in one single product loop βοΈ
overmind truly acts as an ai engineer that can do it all while also continuously improving over time!
Overmind turns anyone into an AI lab.
β’ Builds a context graph from code and traces
β’ Curates training and eval datasets
β’ Evaluates prompts and models
β’ Trains smaller, specialized models
https://t.co/AEBRXmTE6M
you don't need a flamethrower to light a candle. we checked, so you don't have to
and you don't need a frontier model for most agent tasks either. overmind trains one that's built for yours π·
https://t.co/kCIXiRNQnH
Introducing Overmind. The training platform for agents.
Overmind trained models across biomedical, legal, and engineering tasks beating the frontier and achieving:
β’ 7x better accuracy
β’ 20x cheaper usage
β’ 28x less hallucinations
Train your own: https://t.co/AEBRXmUbWk
so what will you actually find here?
mostly me building in public, including the parts that break. notes on ai agents, what's working, what isn't, and what i'm learning along the way. and a lot of travel photos, obviously, plus the occasional adorable doggo gif that never fails to make me smile.
a few things i believe going in. ai keeps making everything faster, but i don't want that speed to cost the joy of actually building things. the hard problems are still the fun part. and the real test for agents isn't another benchmark. it's the day my parents use one without calling me first. that's agi for me.
the process has always mattered more to me than the result, so that's mostly what i'll share. if any of this sounds like you, say hi.
first post on here, so here's a proper intro.
i'm frey. for the last few years i've been an early engineer at ai startups, building them from the ground up, first in san francisco and now in london.
my journey started with not understanding things and being too curious to leave them alone. that led me to parallel computing at @dask_dev, quantum computing at @GoogleQuantumAI, compilers at @LLVMFoundation, ci tooling at @jenkinsci, cloud security at @Oracle, ai bias research at @AvidMldb, ai security at @RealmLabsAI, and now ai agents at @OvermindLab.
outside of work, i'm usually travelling, and the first phrase i learn in every new country is how to ask if i can pet someone's dogπ
okay, that's enough linkedin for one thread.
i love playing cricket and swimming. basketball and volleyball were never really my choice. school coaches started pulling me into both for reasons that were mostly vertical, and people still do.
i'm either on a trip or already planning the next one. i take a borderline concerning number of photos of everything around me, and in every new city, my first search is where to find the best pizza.
the rest goes to league, reading, dancing whenever there's music and enough space, and youtube rabbitholes that start with one video and somehow end three hours later on something i'll never need to know.